Project Summary The overall objective of the present proposal is to test a specific hypothesis about how the developing human brain is able to learn new information from the visual and auditory environment in such an efficient manner during early infancy. Extensive behavioral evidence from infants confirms that they can rapidly learn new combinations of features, but it remains unclear what neural mechanism supports this learning. The hypothesis under examination in the present proposal is based on neural recordings from the visual cortex of developing ferrets, which showed that patterns of activity shifted from being stimulus driven to being predicted by small deviations from background (i.e., non-stimulus driven) activity. That is, the developing ferret brain created a probabilistic model of the most likely features in the environment and used that model as a baseline from which stimulus driven activity was compared. This probabilistic coding model is an efficient way for the brain to represent new visual features because it focuses its activity on the most likely stimuli in the environment and creates patterns of spontaneous activity that are tuned to the environmental mean. The specific aims of the present proposal are to use a newly emerging neuroimaging method, called functional near-infrared spectroscopy (fNIRS), to non-invasively measure the blood oxygenation correlates of neural activity in the visual and auditory regions of the infant brain at four ages: 6 weeks, 3 months, 6 months, and 12 months. Infants will be tested in darkness or silence and in three stimulus conditions in each sensory modality that include both complex features typical of their natural environment and simple features that rarely occur in their natural environment. The probabilistic coding model predicts a gradual progression across post-natal age in the similarity of patterns of neural activity between darkness/silence and natural environmental input, with a corresponding failure to show similarity between darkness/silence and the non-natural stimulus conditions. Should the probabilistic coding model be supported, it would enable assessments of infants from at-risk or special populations, such as Autism Spectrum Disorder, both to establish an early biomarker of brain disorders and to serve as a possible explanation for what property of the neural system is aberrant in these disorders.